As semiconductor memory devices, dynamic RAMs (DRAMs) formed using a silicon substrate are well-known products and currently used in a variety of electronic devices. A memory cell which is a key component in a DRAM includes a reading and writing transistor and a capacitor.
DRAMs are an example of volatile memory devices, and another example of volatile memory devices is static random access memories (SRAMs). An SRAM holds stored data by using a circuit such as a flip-flop and thus does not need refresh operation, which is an advantage over a DRAM. However, cost per storage capacity is high because a circuit such as a flip-flop is used. Moreover, as in a DRAM, stored data in an SRAM is lost when power supply stops.
An example of nonvolatile memory devices is flash memories. A flash memory includes a floating gate between a gate electrode and a channel formation region in a transistor and stores data by holding charge in the floating gate. Therefore, a flash memory has advantages in that the data holding period is extremely long (semi-permanent) and refresh operation which is necessary to volatile memory devices is not needed.
However, in a flash memory, there is a problem in that a memory element becomes incapable of functioning after a large number of repeated write operations because a gate insulating layer included in the memory element deteriorates due to tunneling current generated in the write operations. In order to avoid this problem, a method in which the number of write operations is equalized among memory elements can be employed, for example, but a complex peripheral circuit is needed to realize this method. Moreover, even when such a method is employed, the fundamental problem of lifetime cannot be resolved. In other words, a flash memory is not suitable for applications in which data is frequently rewritten.
In addition, high voltage is necessary in order to inject charge into the floating gate or remove the charge. Moreover, it takes a relatively long time to inject or remove charge, and it is not easy to increase the speed of writing or erasing data.
Volatile memory devices typified by DRAMs are formed using a silicon substrate, and their circuit patterns, like those for other semiconductor integrated circuits, have been miniaturized in accordance with the scaling law. There was a time when it was considered difficult to achieve a design rule of 100 nm or less. One of the reasons is that in a transistor having a channel length of 100 nm or less, a punch-through current is likely to flow due to a short-channel effect and the transistor becomes incapable of functioning as a switching element, which has been considered to be a problem. In order to prevent a punch-through current, a silicon substrate may be doped with an impurity at high concentration. However, this is not an appropriate solution to the problem because it makes a junction leakage current likely to flow between a source and the substrate or between a drain and the substrate and eventually causes a deterioration of memory retention characteristics.
Reference 1 discloses a technique in which a transistor formed using an oxide semiconductor layer is used as a memory.
Reference 2 discloses a transistor formed using amorphous IGZO and having a channel length of 50 nm.